| can anybody explain to me the term regularization within
| the defaults/spatial normalisation area?
Spatial normalisation minimises two cost functions: one based
on the sum of squared difference between the image and template(s),
and the other based on maximising the smoothness of the deformations.
Regularisation involves stabilising the estimation of the deformations
by penalising large deformations.
This can be thought of in a Bayesian framework, as:
P(warp|data) \propto P(data|warp)*P(warp)
i.e., the probability of a deformation given the data is proportional
to the likelihood of observing the data given a deformation field
times the prior probability of the deformation field. For spatial
normalisation, P(data|warp) is related to the sum of squares difference
between the images, whereas P(warp) is related to the smoothness cost
function.
Best regards,
-John
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|